Exploratory Analysis of Red Wine Quality by Venkat Ramesh
The Red Wine Quality dataset contains quality ratings on a scale of 0 to 10, accompanied by differetnt attributes of red wine.
Below is a snapshot of the various columns in the dataset and their datatypes:
## [1] 1599 13## 'data.frame': 1599 obs. of 13 variables:
## $ X : int 1 2 3 4 5 6 7 8 9 10 ...
## $ fixed.acidity : num 7.4 7.8 7.8 11.2 7.4 7.4 7.9 7.3 7.8 7.5 ...
## $ volatile.acidity : num 0.7 0.88 0.76 0.28 0.7 0.66 0.6 0.65 0.58 0.5 ...
## $ citric.acid : num 0 0 0.04 0.56 0 0 0.06 0 0.02 0.36 ...
## $ residual.sugar : num 1.9 2.6 2.3 1.9 1.9 1.8 1.6 1.2 2 6.1 ...
## $ chlorides : num 0.076 0.098 0.092 0.075 0.076 0.075 0.069 0.065 0.073 0.071 ...
## $ free.sulfur.dioxide : num 11 25 15 17 11 13 15 15 9 17 ...
## $ total.sulfur.dioxide: num 34 67 54 60 34 40 59 21 18 102 ...
## $ density : num 0.998 0.997 0.997 0.998 0.998 ...
## $ pH : num 3.51 3.2 3.26 3.16 3.51 3.51 3.3 3.39 3.36 3.35 ...
## $ sulphates : num 0.56 0.68 0.65 0.58 0.56 0.56 0.46 0.47 0.57 0.8 ...
## $ alcohol : num 9.4 9.8 9.8 9.8 9.4 9.4 9.4 10 9.5 10.5 ...
## $ quality : int 5 5 5 6 5 5 5 7 7 5 ...Univariate Plots Section
Let’s explore each of the columns individually, to begin with.
Quality
Quality is the label column of interest that contains the integer score between 0 and 10. Here’s the histogram of quality ratings in the dataset.
We see a nice normal distribution with most of the quality scores centered around 5. Subsetting the data into those with quality less than 5 and greater than 5 will help us see trends, since it also makes sense intuitively to place the pass score for quality at, greater than 5. We also notice that this splits our data into roughly two equal halves.
##
## low_quality high_quality
## 744 855When we look at our attributes, we will also generate a faceted version of the attribute histogram to get a peek ahead on the interaction of this attribute to th quality. Since all of the columns are of decimal datatype, we will pick 0.1 as our binwidth to be able to capture with enough granularity, unless we encounter a case where the range is too big or two small for this value.
If the distribution deviates significantly from normal distribution and is long tailed, we will apply log10 transformation to the x axis to get a better picture.
Alcohol
We see that low-quality group has long-tailed distribution for alcohol levels, whereas the high-quality group appears to be more uniform. We will revisit this to investigate the correlation between quality and alcohol in our bivariate analysis.
Sulphates
The distribution seems right skewed, but log10 transformation shows nothing different in the high quality vs low quality plots.
pH
The distribution looks normal so no transformation is needed.
Density
Apart from the fact that the long tail outliers on the right of the distribution belong to the high quality dataset, this is a again pretty normal.
Total Sulfur Dioxide
We see from the data description that total sulfur dioxide of greather than 50ppm, becomes evident in taste. We need to pay attention to see if this has an effect on perceived quality.
Free Sulfur Dioxide
We would expect free sulfur dioxide to be correlated with total sulfur dioxide. We will examine this in the bi-variate section.
Chlorides
The distributions for cholrides are identical in both quality classes except for the fact that the peak around mean is higher for high quality subset.
Residual Sugar
We notice that the long tail on the right of high quality subset, and would like to investigate if residual sugar has an effect on perceived quality. Also, in general, the distribution of residual sugar has considerably far-off outliers as evidenced in the box plot below.
Citric Acid
We notice an almost trimodal distribution for citric acid in the low quality case, and they occur around means of 0.0, 0.25, 0.5 etc.
Volatile acidity
Fixed Acidity
Univariate Analysis
What is the structure of your dataset?
There are 1599 rows of wine quality data. Each row has 12 columns of numerical attributes describing aspects of the red wine and a column of integers that corresponds to a quality score between 0 and 10.
What is/are the main feature(s) of interest in your dataset?
Upon looking at Univariate plots, the following column appear to be interesting in understanding the main feature of interest which is the quality of wine. - Higher alcohol seems to be associated with high quality wine subset. - Suphur dioxide dependence and correlation between total and free levels needs to be looked at. - Acidity related columns such as pH, Citric Acid, fixed and volatile acidity need to be looked at overall acidic wines with lower pH seem to be higher quality .
What other features in the dataset do you think will help support your
investigation into your feature(s) of interest?
N/A
Did you create any new variables from existing variables in the dataset?
We generated a quality class based on whether the quality score is greater than 5 or not. We may have to create additional factor variables from numericals for trivariate analysis.
Of the features you investigated, were there any unusual distributions?
Did you perform any operations on the data to tidy, adjust, or change the form
of the data? If so, why did you do this?
The distributions that were right skewed were transformed using log10 scaling.
Bivariate Plots Section
Let’s start our bivariate analysis by analyzing key columns identified earlier for their effects on quality.
Quality vs Alcohol
The mean alcohol level at each quality level is denoted by ‘x’. We see that overall higher alcohol levels are asssociated with higher quality, and at the highest quality all the datapoints are above 7 % alcohol.
Quality vs Total Sulfur Dioxide
The total sulfur dioxide relation is more complicated. At low levels of less than 50ppm total sulfur dioxide, the datapoints are distributed across all quality levels whereas the datapoints with greater than 120ppm or so, are invariably in the 5-6 range always, indicating that at higher levels of sulfur dioxide, its negative impact on perceived quality is more pronounced.
Quality vs pH
Lower pH seems to be associated with higher quality indicating acidic wines are better tasting.
Quality vs Volatile Acidity
However, volatile acidity is not favored since higher volatile acidity leads to lower wine quality. Beyond volatile acidity value of 1, there are no more high quality wine datapoints.
Quality vs Residual Sugar
Residual sugar doesn’t seem to be indicative of quality, which is contrary to intuition. Also, the IQR of residual sugar in the dataset is small, with some far off outliers which seem associated with low and mid-quality wines mostly.
Scatter matrix
Now, let’s look at the scatter matrix to identify correlation among attributes.
As expected, the acidity metrics are highly correlated with one another. We also see correlations between acidity metrics and density, and between density and alcohol content. As expected, the sulfur dioxide levels also correlate well for free vs total levels.
Free vs Total Sulfur Dioxide
The points spread out more at higher total levels and overall it appears as though more free to total ratio might result in higher quality.
Fixed acidity vs pH
There is a strong negative correlation as lower pH levels mean more acidity.
Fixed vs Volatile Acidity
Again, we can see from this plot that high quality wines seem to have lower volatile to fixed acidity ratios.
Fixed acidity vs Citric Acid
Addition of higher levels of citric acid seem to be associated both with higher quality and also higher fixed acidity.
Density vs Fixed Acidity
High quality wine seems to be associated with lower density for given fixed acidity.
Density vs Alcohol
As expected most of the high quality wines are associated with higher alcohol levels and higher alcohol levels also seem to correlate with lower density.
Bivariate Analysis
Talk about some of the relationships you observed in this part of the
investigation. How did the feature(s) of interest vary with other features in
the dataset?
High quality wine seems to be mosly acidic (lower pH) but with low levels of volatile acidity. Alcohol levels are directly proportional to perceived quality. High quality wines also seem to be less dense.
Did you observe any interesting relationships between the other features
(not the main feature(s) of interest)?
Density and fixed acidity seem to be strongly related in a directly proportional way.Density and alcohol content are inversely proportional.
What was the strongest relationship you found?
Higher alcohol levels seem to be associated most strongly with higher quality. Similarly, lower pH is also associated with higher quality.
Multivariate Plots Section
Let’s extend the visualization of relation between density and alcohol from previous section also adding a residual sugar class (less than median is low sugar and greater than median is high sugar.)
Sweeter wines with higher sugar are also associated with greater density. However, higher alcohol levels are associated with lower density.
This plot show further shows the effect of density and fixed acidity on quality. The darker colors are invariable on the lower envelope of points showing that for a given fixed acidity, lighter density wines are higher quality.
In this plot, the darker points are mostly at the top, showing the effects of two of the strongest indicators of quality that we identified, namely alcohol and acidity levels. This shows that hight quality wines are mostly high on alcohol and occur on lower pH levels.
This shows a positive indicator of quality i.e. alcohol vs negative indicator of quality namely volatile acidity and clearly the high quality wines are high on alcohol and low on volatile acidity.
Multivariate Analysis
Talk about some of the relationships you observed in this part of the
investigation. Were there features that strengthened each other in terms of
looking at your feature(s) of interest?
Lower pH levels, lower volatile acidity and higher alcohol levels strengthened each other in improving wine quality.
Were there any interesting or surprising interactions between features?
Surprisingly residual sugar did not play a role in quality determination even though it increased density thereby potentially lowering quality rating.
OPTIONAL: Did you create any models with your dataset? Discuss the
strengths and limitations of your model.
N/A
Final Plots and Summary
Plot One
This plot is interesting because it explains the relationship of sugar and alcohol to density which was alluded to in wineQualityInfo.txt. Higher sugar wines seem to have a consistent lead in density compared to wines in lower half in residual sugar range. Alcohol on the other hand, brings down the density at higher levels. This also runs counter to intuition that sweeter wines would be better received, since we see that less dense and higher alcohol level wines are normally preferred in this dataset.
Plot Two
This graph succintly captures the degradation in wine quality with increasing volatile acidity.
Plot Three
This final plot shows the remaining pieces of the trend in high quality wines of having higher alcohol levels and lower pH levels indicating that acidic and more concentrated wines are preferred.
Reflection
The biggest challenge in dealing with this dataset was the lack of domain knowledge on the account of being a teetotaler. However, exploring one variable at a time and faceting by quality class to identify candidates for two variable and then faceting again during two variable analysis to identify three variable interactions, seem to uncover the relationships organically. Scatter matrix was particulary helpful as were summary stats in discovering trend.